Algorithm Of 3D Hand Posture Recognition With Space Coordinates Based On Optimal Feature Selection

As an important branch of human-computer interaction,gesture recognition has been widely used in production and living,which is of great significance to improving the way of working and promoting the qualities of life.It has great value to recognize hand posture since people with hearing and speech disabilities use sign language as the main medium of communication,which prevents them from communicating with the normal.In gesture recognition,several elements are included in the gesture which are hand shape,hand position and movement.And hand shape which represents the shape of gestures performed is the most intuitive description of gestures among them.Therefore,it is crucial for hand gesture recognition to recognize hand posture(i.e.,hand shape).The summary of current researches at home and abroad is given in this paper,and the problems of existing methods are analyzed then.There are three directions of hand gesture recognition,which are 2D hand gesture recognition based on 2D images,3D hand gesture recognition based on depth images and 3D hand gesture recognition based on space coordinates.At present,there are few researches of the last direction and the recognition result is affected by redundant features in those papers without feature comparison and selection.In addition,the recognition result is also influenced by the direction of hand gestures performed.In order to eliminate the shortcoming of low recognition rate caused by the redundant features and improve the problem brought by the direction of hand gestures performed,an algorithm of 3D hand posture recognition based on optimal feature selection is proposed in this paper.The four main steps included in the recognition process are data acquisition,feature extraction,optimal feature selection and posture recognition.Firstly,3D coordinate data of key points from hand are collected by Leap Motion Controller.Secondly,self-defined attributes and features are extracted from them.Thirdly,the XGBoost algorithm combined with cross validation is employed to select optimal features from different attributes.Finally,the selected features instead of all extracted features are then fed into Gaussian Naive Bayes classifier to recognize the target posture.The proposed method is experimented on two testing sets containing different data sequences of ten heavily-used postures of Chinese Sign Language,which are postures with direction and postures without direction.The experimental results show that after processed by optimal feature selection,the proposed method can achieve higher recognition rate than the traditional methods in both two testing sets.In the testing set with direction,the recognition rate is almost 100%,and the number of training samples needed less than other methods to the peak recognition rate.In the testing set without direction,the recognition rate can reach to 95%.In this paper,the machine learning and data mining are combined innovatively to realize the three-dimensional gesture recognition.The proposed method not only has high recognition rate and good robustness,but can also reduce the collection of 3D coordinate of key joints of postures,which reduces the error brought by the other joints since they will be sheltered from other part of hand.